The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external REST, SOAP, and SFTP sources through various callbacks and client methods. Processing untrusted external content inherently presents an attack surface for indirect prompt injection if the agent interprets that data as instructions. Ingestion points: parseResponse in HTTP-SERVICES.md, SOAP-SERVICES.md, and FTP-SERVICES.md; SFTPClient.get() in FTP-SERVICES.md. Boundary markers: Not explicitly defined in the example service callbacks or prompt templates. Capability inventory: Network operations via LocalServiceRegistry and file system access within the sandbox via dw.io.File. Sanitization: The skill recommends filterLogMessage for logs to protect outgoing data, but lacks comprehensive input sanitization guidelines for data received from external services.

b2c-webservices